Backtesting some robo-advisers asset allocations

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 7:13 pm

An interesting thread about robo-advisers has been recently unfolding, leading to the creation of a corresponding wiki page.

Somewhere in there (this post), it was suggested to backtest some recommendations from robo-advisers, starting by the "Schwab Intelligent Portfolio" (aka SIP), and possibly others.

Such backtesting would be performed by approximating the definition of the portfolio in the Bogleheads Simba spreadsheet, then assume a fixed asset allocation (not all robo-advisers do that though), and see what goes.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 7:16 pm

jbolden1517 provided a great description of a SIP sample portfolio, let me quote it for reference.
jbolden1517 wrote:Schwab Intelligent Portfolios

Schwab's doesn't charge for their robo-advisor service. Fees for funds average about 17 basis points. They also unlike most robo-advisors have a pronounced value tilt. They give this sample portfolio of a 2/3rds stock 1/3rd bond
Image

Fundamental are Schwab value ETFs using 3 value factors (Sales weighting, cash flow weighting and dividend weighting) (more details: https://intelligent.schwab.com/public/i ... -etfs.html). The ones without Fundamental are standard indexes (market capitalization weighting). The stock allocation has a solid value and small tilt. Most of the bond allocation (15%) is higher risk, high return brings down correlation a bit. It should be noted this is a substantial deviation from the 3-fund style portfolios where bonds are used for safety. These bonds include both duration risk and credit and are meaningfully generating return. In addition to the duration and credit risk on the bonds component this fund uses the equity as a core and introduces 2 main diversifiers. The diversifiers help withnon-correlation (non-correlation is important for increasing serial returns) while retaining a low fees (i.e. substantial non-correlation can be expensive to implement). Previous metals traditionally do well in crisis and inflation. The 6% in low risk bonds similarly is a diversifier that will do well especially against deflationary bears. They along with the cash are designed to give the investor something to rebalance out of.

Schwab offers a detailed list of the funds they use to construct the portfolio and how they are expected to perform: https://intelligent.schwab.com/public/i ... asses.html

The funds are offered in pairs because Schwab offer tax loss harvesting for portfolios over $50k using fund substitution.

They publish quarterly updates on how the components do (example 2Q2017): https://intelligent.schwab.com/public/i ... eview.html

Schwab is recognized as one of the best discount brokers for customer service. Schwab as a brokerage has strong banking features and sells a range of annuities... and these can be coordinated with the robo-advisor service (though it won't rebalance around them). The annuity feature is particularly important for income oriented retirees. The robo-advising can be used for revocable living trusts accounts which is a differentiating feature few other robos offer. The Schwab goal-tracker estimates how close the savings component is given ranges of portfolio returns to achieving an investment goal for a future timeframe (https://intelligent.schwab.com/public/i ... /goal.html)

For human guidence Schwab offers an upgrade to a human advisor for .28% with a cap at $900/ quarter (comparatively cheap). They also offer what they claim is full wealth management at .7-.9% with at $500k and .3% at $10m. This is a low min, lowest fee I've seen for wealth management, though the range of services also seems more middle class tailored (the sorts that a financial advisor would perform) than I've seen for wealth management. My gut is that this is gimmicky title aimed at people who don't actually need wealth management.

At this point Schwab Intelligent Portfolios is my recommended robo-advisor, and recommended simple portfolio.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 7:23 pm

First, let's try to map those various asset classes to the asset classes for which we do have significant history in the Simba spreadsheet. The red mappings are very coarse and somewhat arbitrary (but those are about small numbers, and differences shouldn't be terribly consequential).

The biggest thing are the blue numbers, as SIP uses 'fundamental' ETFs, which basically means 'smart beta', and that I equated to 'value' as defined by Vanguard and the indices it follows. Some would argue this is a poor mapping, ignoring the 'fundamental' premium and might make the entire exercise moot... Others would argue that smart beta is clever marketing for what was known as value for the longest time... and... I don't want to argue, this is simply the best we can do with Simba! Also note that all portfolios in the comparison to come are judged by the same standard, i.e. the Vanguard definition of 'value'.

See the mapping table below. I am happy to tweak it for another experiment as needs be. Click on the image for a larger display.

Image

EDIT: fixed the cash allocation to money market
Last edited by siamond on Wed Aug 02, 2017 7:34 pm, edited 1 time in total.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 7:47 pm

As you can notice in the previous matrix, the SIP portfolio is 70% stocks (approximating precious metals to stocks) and 30% bonds (approximating cash to bonds), and the stocks are 50/50 between US and International.

The most obvious comparison is therefore a 3-funds portfolio with the same split (35/35/30). It was also requested to compare to a 80/20 3-funds portfolio - not too sure why, but it's easy, so I added it.

But then a good number of Bogleheads do not follow this kind of purist approach of a 3-funds portfolio, and they would tune the allocation by adding various tilts. So I quickly designed two additional portfolios, derived from the 70/30 pattern, with moderate tilts in one case, and more aggressive tilts in the second case. While keeping bonds very simple and US only as the majority of Bogleheads tend to do that, tilters or not tilters. Sure, I made various arbitrary calls, I wasn't trying to optimize one aspect or another, I was just trying to just come up with something simple and fairly representative of what a Bogleheads might do. As a side note, the aggressive tilt has quite some commonalities with my own portfolio (although I don't use 30% bonds to begin with!).

Image

EDIT: fixed the cash allocation to money market
Last edited by siamond on Wed Aug 02, 2017 7:38 pm, edited 1 time in total.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 7:58 pm

Let's start by the usual Risk (defined as volatility/std-deviation) vs. Reward (defined annual returns/CAGR) chart. This applies to $100 invested on 31-Dec-1984, and the subsequent trajectory from 1985 to 2016 (we don't have the historical returns for several of the asset classes involved before 1985). Assuming annual rebalancing and a fixed asset allocation.

Image

For people who do agree with such definition of risk (as the academics tend to do, while real life might be viewed with different lens), then the 80/20 portfolio didn't fare very well. Between the SIP portfolio, the 70/30 3-funds portfolio and the moderate tilt portfolio, well, it depends on one's perspective (the moderate tilt has a slightly better Sharpe ratio - see next post). The aggressive-tilt portfolio did fare better.

I would observe that is the 9% cash is quite a drag for the SIP portfolio, and the subtleties with the various types of bonds may not be entirely effective.

EDIT: fixed the cash allocation to money market
Last edited by siamond on Wed Aug 02, 2017 7:46 pm, edited 1 time in total.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 8:03 pm

For those of you who want more numbers, here are a bunch of metrics. Let me refer you to the Simba spreadsheet itself for explanations on the various metrics being displayed. You can also see that there is conditional formatting on the various rows, to ease the visualization of rankings for a given metric.

Image

EDIT: fixed the cash allocation to money market
Last edited by siamond on Wed Aug 02, 2017 7:47 pm, edited 1 time in total.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 8:12 pm

Personally, I find such numbers a little dry, and I became very fond of telltale charts (this wonderful tool created by Jack Bogle himself). A telltale chart shows the aggregate growth of a given portfolio (or fund) divided by the aggregate growth of a reference portfolio (or fund). This really helps in showing the dynamics year over year, in a much more effective way than a growth chart.

Here, the 70/30 3-funds portfolio is used as a reference. And all 5 portfolios are displayed on the telltale. The SIP portfolio got a rough start (again, relative to a 70/30 3-funds portfolio), suffered dearly from the Internet crisis (like many other portfolios), spectacularly recovered (cf. the big jump of the value stocks in 2002+), and then fell pretty hard during the financial crisis and didn't do too well since then. While the tilted portfolios behaved a tad better imho if... you didn't get a heart attack around 2000 and stayed the course (easy to say, much harder to do!).

Image

EDIT: fixed the cash allocation to money market
Last edited by siamond on Wed Aug 02, 2017 7:52 pm, edited 1 time in total.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 8:14 pm

I also ran all 10-years periods between 1985 and 2016 via a data table, and computed average metrics for those periods (e.g. CAGR, Sharpe, Sortino, etc). The outcome is basically the same though, so I will only post the corresponding table if requested.

Now let me open the door for feedback. Does this analysis make sense? It is certainly approximate, in all fairness. Do you perceive the numbers and charts in a different way than I did? Etc.

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Tue Aug 01, 2017 8:43 pm

Nice analysis thank you. The reason we were asking for 70/30 and 80/20 is my gut was that SIP 70/30 would have a risk somewhere in that range, and well looking at the graph seems like I was right. Not asking you to do all the posting over but SIP is 60/40 USA/Int not 50/50 like the other portfolios you tested against. I do have some serious concerns about the accuracy of proxying a sales/cashflow/dividend weighted fund with a cap weighted value metric fund. I get that we don't have the data on this sort of fund, but it is something to keep in mind that the returns between this sort of weighting and cap weighing might not be similar enough. I figure we will know more after the next bear market to see how well the 2 types of value tracked through a full cycle.

As far as the comparison with 3-fund. IMHO Tilt1 and Tilt2 IMHO are nothing like 3 fund, those are more than tilts they are entirely different asset allocation. I think Tilt2 demonstrates that the more you move away from large cap growth and into value the better.

The one thing that is interesting about SIP is how lousy the fund was 1984-1990. That's the period that did the damage, after that it keeps up with Tilt2 (the clear winner). I would have thought the high yield in the mid 1980s would have been a huge performance kick. Could you let me know what's happening in that period that's causing the underperformance?

BTW where do you get that Bogle charting tool?

User avatar
LadyGeek
Site Admin
Posts: 40469
Joined: Sat Dec 20, 2008 5:34 pm
Location: Philadelphia
Contact:

Re: Backtesting some robo-advisers asset allocations

Post by LadyGeek » Tue Aug 01, 2017 9:27 pm

I agree, that's a nice analysis. Thanks.

The telltale chart was intended to show the effects of Mean reversion, but I like the implementation for backtesting.

I think the telltale charts support an important point that may be lost in the noise of the details -

Robo-adviser portfolios may sometimes do better, sometimes do worse than a "conventional / DIY" Lazy portfolio. However... If you know absolutely nothing about investing, a robo-adviser portfolio will do its best to preserve your savings for retirement.

Of course, you could override the default settings and it will tank, but you can do that with any portfolio.

On a side note, I think Target date funds have been providing "automated" portfolios for quite some time. There's still nothing wrong with selecting a "set and forget" fund. Again, if you don't know what you're doing, they're just fine.
jbolden1517 wrote:BTW where do you get that Bogle charting tool?
It's part of the backtesting spreadsheet, see: Simba's backtesting spreadsheet

siamond - could you please add the definition of telltale chart in the glossary? Perhaps a definition like:

Telltale chart - A telltale chart shows the aggregate growth of a given portfolio (or fund) divided by the aggregate growth of a reference portfolio (or fund). This really helps in showing the dynamics year over year, in a much more effective way than a growth chart. Reference: The Telltale Chart, June 26, 2002.
To some, the glass is half full. To others, the glass is half empty. To an engineer, it's twice the size it needs to be.

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Tue Aug 01, 2017 9:34 pm

LadyGeek wrote: On a side note, I think Target date funds have been providing "automated" portfolios for quite some time. There's still nothing wrong with selecting a "set and forget" fund. Again, if you don't know what you're doing, they're just fine.
I agree. As well as global allocation funds that offer low volatility (and thus safe draws for income and good compounding without the need to externally rebalance). In many ways those sorts of funds keep with John Bogle's idea that mutual funds should be a middle class investment vehicle designed for investing + holding for a lifetime.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 10:52 pm

jbolden1517 wrote:Nice analysis thank you. The reason we were asking for 70/30 and 80/20 is my gut was that SIP 70/30 would have a risk somewhere in that range, and well looking at the graph seems like I was right. Not asking you to do all the posting over but SIP is 60/40 USA/Int not 50/50 like the other portfolios you tested against.
I hesitated a little bit about this one, but no, I believe this is 50/50 USA/Int'l. Just exclude precious metals, do the math, this is indeed 50/50. As to precious metals, well, when looking at VGPMX, 65% of it is actually in Canada, and only 10% is in the USA (check here).... As to a risk-adjusted perspective, well, this is what the Sharpe ratio is about, and I was actually surprised to see that SIP and 3-funds 70/30 were equivalent (a tiny advantage to SIP when looking at all decades, actually, but very small).
jbolden1517 wrote:I do have some serious concerns about the accuracy of proxying a sales/cashflow/dividend weighted fund with a cap weighted value metric fund. I get that we don't have the data on this sort of fund, but it is something to keep in mind that the returns between this sort of weighting and cap weighing might not be similar enough. I figure we will know more after the next bear market to see how well the 2 types of value tracked through a full cycle.
Yes, I hear you, this was my far my biggest approximation. I've heard both sides of the coin arguing until they were blue on the face, but fact is... there is just not enough historical data available on this to establish a sound judgement on this. Future will tell (might have to wait 50 years though!). I suspect the smart-beta algorithms will change as times go by though... :wink:
jbolden1517 wrote:As far as the comparison with 3-fund. IMHO Tilt1 and Tilt2 IMHO are nothing like 3 fund, those are more than tilts they are entirely different asset allocation. I think Tilt2 demonstrates that the more you move away from large cap growth and into value the better.
I don't quite agree. To build Tilt1, I just took some % out of TSM and sprinkled it to overload SCV and REITs (and same for Int'l). This is truly how Bogleheads tilters do it, as far as I understand. Tilt2 is more aggressive for sure, and less people would go there, although the historical evidence is indeed very strong. Now will the future rhyme with the past, we shall see...
jbolden1517 wrote:The one thing that is interesting about SIP is how lousy the fund was 1984-1990. That's the period that did the damage, after that it keeps up with Tilt2 (the clear winner). I would have thought the high yield in the mid 1980s would have been a huge performance kick. Could you let me know what's happening in that period that's causing the underperformance?
The drag from the 9% cash is is the main issue, just moving it to bonds basically fixes the portfolio and brings it back to par with my tilting portfolios. Precious metals also hurt a bit. I suppose SIP uses a fairly extreme rebalancing technique requiring such amount of cash (as opposed to the simple annual rebalancing I assumed), but I've run enough simulations of rebalancing algorithms to be very skeptical about their ability to generate a true premium.

IMPORTANT EDIT: just realized that I was too literal in my interpretation of cash. I interpreted cash as hard cash, something returning 0% a year. But coming back to Schwab's guide of the funds they use, of course, they do the obvious and use a money-market fund. So I should have mapped the 9% cash to something like VUSXX in my backtesting. I just gave it a quick try, and this definitely helps make SIP better. Still not as good as simply put those 9% in Total-Bonds, but this does help.
Last edited by siamond on Tue Aug 01, 2017 11:55 pm, edited 1 time in total.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 11:00 pm

jbolden1517 wrote:
LadyGeek wrote: On a side note, I think Target date funds have been providing "automated" portfolios for quite some time. There's still nothing wrong with selecting a "set and forget" fund. Again, if you don't know what you're doing, they're just fine.
I agree. As well as global allocation funds that offer low volatility (and thus safe draws for income and good compounding without the need to externally rebalance). In many ways those sorts of funds keep with John Bogle's idea that mutual funds should be a middle class investment vehicle designed for investing + holding for a lifetime.
Personally, I am very impressed by the Life Strategy funds. Those are the ultimate 'set and forget' funds imho. The target-date funds going down to 30/70 are just a disaster in the making if you ask me (they definitely backtest VERY poorly).

I do agree with you guys that it is crucial to have such 'one fund' offerings (or one robo-advisor acting as a -cheap- 'mutual fund manager' of sorts), which are really the best option for the great majority of people who just don't want to touch investment-related thinking with a long pole and yet do need to retire at some point...

Anyhoo, back to backtesting, if you want me to test other robo-advisers portfolios (or some variations of SIP), don't hesitate to ask. It really doesn't take me long.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Tue Aug 01, 2017 11:21 pm

Here is the customized spreadsheet I used to perform this analysis. If somebody wants to play around, feel free to make a copy and have at it... Go to Compare_Portfolios to find the various tables and charts I posted. Feel free to ask questions.

https://drive.google.com/open?id=0B0svR ... ndib0k4cUk

billthecat
Posts: 102
Joined: Tue Jan 24, 2017 2:50 pm

Re: Backtesting some robo-advisers asset allocations

Post by billthecat » Tue Aug 01, 2017 11:43 pm

siamond wrote:
jbolden1517 wrote:
LadyGeek wrote: On a side note, I think Target date funds have been providing "automated" portfolios for quite some time. There's still nothing wrong with selecting a "set and forget" fund. Again, if you don't know what you're doing, they're just fine.
I agree. As well as global allocation funds that offer low volatility (and thus safe draws for income and good compounding without the need to externally rebalance). In many ways those sorts of funds keep with John Bogle's idea that mutual funds should be a middle class investment vehicle designed for investing + holding for a lifetime.
Personally, I am very impressed by the Life Strategy funds. Those are the ultimate 'set and forget' funds imho. The target-date funds going down to 30/70 are just a disaster in the making if you ask me (they definitely backtest VERY poorly).
How do you backtest a fund that changes over time?

Why do they backtest poorly given the conventional wisdom of slowly moving to a higher bond position to reduce risk / volatility?

User avatar
in_reality
Posts: 4219
Joined: Fri Jul 12, 2013 6:13 am

Re: Backtesting some robo-advisers asset allocations

Post by in_reality » Tue Aug 01, 2017 11:59 pm

siamond wrote: The drag from the 9% cash is is the main issue, just moving it to bonds basically fixes the portfolio and brings it back to par with my tilting portfolios. Precious metals also hurt a bit. I suppose SIP uses a fairly extreme rebalancing technique requiring such amount of cash (as opposed to the simple annual rebalancing I assumed), but I've run enough simulations of rebalancing algorithms to be very skeptical about their ability to generate a true premium.
I'm not sure you are really fully capturing the SIP returnsh.

First, why is cash at 0%? Schwab pays according to the average money market return. Vanguard has the 10 year Money Market Funds Average 0.47%. see https://www.rate-watch.com/national-averages

Also, why is it important to break US stocks down into small value, but not important for international? OK lack of data. But SIP does give small value international developed exposure. It also gives emerging value exposure (large and mid-cap too).

OK so EFV is your proxy. Well here are actually returns Jan 2009 - Jul 2017.

Fund / CAGR / STDEV
EFV / 7.14% / 19.09%
*SFILX / 13.12% / 16.98%
*SFL / 8.66% / 19.47%

*SFS Schwab Fundamental Small Cap International
*SFL Schwab Fundamental Large Cap International

Schwab Fundamental Small Cap International has had about a 2% higher CAGR during the life of Vanguard FTSE All-World Ex-US Small-Cap Index Fund Investor Shs (VFSVX).

Schwab Fundamental Emerging has had about a 1% lower CAGR compared to VEIEX (over Schwab's lifetime).

Schwab Fundamental US Lg Company Index Fund Insti Shs had about a 1.5% higher CAGR than VIVAX Vanguard Value Index Fund (over the Schwab fund's lifetime).

Anyway, why did SIP do so poorly in backtesting '85-'90?
Last edited by in_reality on Wed Aug 02, 2017 12:06 am, edited 1 time in total.
[60% US _ 26% DEV _ 14% EM] | (-16% LC _ +8% MC _ +8% SC) | [47% FND/VAL _ 40% MKT _ 7% MOM _ 6% REIT] | (+/- 5% or *25% rebalancing bands)

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Wed Aug 02, 2017 12:04 am

billthecat wrote:
siamond wrote:Personally, I am very impressed by the Life Strategy funds. Those are the ultimate 'set and forget' funds imho. The target-date funds going down to 30/70 are just a disaster in the making if you ask me (they definitely backtest VERY poorly).
How do you backtest a fund that changes over time?
Not with the Simba spreadsheet, which is hard-wired for fixed allocations. But with a custom spreadsheet, this isn't hard to do. I don't have anything ready to be shared publicly though, I was referring to private experiments of mine (and others).
billthecat wrote:Why do they backtest poorly given the conventional wisdom of slowly moving to a higher bond position to reduce risk / volatility?
Well, you will find that quite a few people on this board would challenge this 'conventional wisdom', and would much prefer staying with a fixed asset allocation for good. To be fair, the 'conventional wisdom' is not about cold hard numbers. It's really about behavior and psychology. It kind of makes the assumption that older people may have enough money to allow less stocks in their portfolio, while not necessarily be able (or willing) to deal with high(er) volatility any more. If those assumptions are true, then why not reducing your stock exposure. But you really don't have to...

This being said, going to 60/40 or 50/50 is one thing (and that's perfectly reasonable for many people), while going to 30/70 is another (that is quite extreme, and can really backfire after a couple of decades of retirement - this is what I was alluding to).

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Wed Aug 02, 2017 12:17 am

in_reality wrote:First, why is cash at 0%? Schwab pays according to the average money market return. Vanguard has the 10 year Money Market Funds Average 0.47%. see https://www.rate-watch.com/national-averages
Yes, you are absolutely correct, my bad. I actually realized my mistake 30 minutes ago - see the EDIT to my answer to jbolden. SIP still doesn't fare very well when I correct this mistake, but certainly better than in my first test. This is why I asked for feedback! :wink:
in_reality wrote:Also, why is it important to break US stocks down into small value, but not important for international? OK lack of data. But SIP does give small value international developed exposure. It also gives emerging value exposure (large and mid-cap too).
And there is no such historical data that I am aware of. I made clear in this post that I had to make a few choices like that. I seriously doubt this changes the big picture though.
in_reality wrote:OK so EFV is your proxy. Well here are actually returns Jan 2009 - Jul 2017.

Fund / CAGR / STDEV
EFV / 7.14% / 19.09%
*SFILX / 13.12% / 16.98%
*SFL / 8.66% / 19.47%

*SFS Schwab Fundamental Small Cap International
*SFL Schwab Fundamental Large Cap International

Schwab Fundamental Small Cap International has had about a 2% higher CAGR during the life of Vanguard FTSE All-World Ex-US Small-Cap Index Fund Investor Shs (VFSVX).

Schwab Fundamental Emerging has had about a 1% lower CAGR compared to VEIEX (over Schwab's lifetime).
Ok, so I guess you're a smart beta believer. Fair enough, I respect that. Please appreciate that eight years of over performance doesn't demonstrate much though (heck, small value underperformed TSM for *decades* in the past). I am not saying you're wrong, mind you, I am just saying we don't have the proper historical data to assess the past on a significant enough time period (i.e. several decades), nor to backtest. Plus new strategies have this tendency of doing well to begin with, then falter. Future will tell. Your point is well taken, though.
Last edited by siamond on Wed Aug 02, 2017 12:22 am, edited 1 time in total.

billthecat
Posts: 102
Joined: Tue Jan 24, 2017 2:50 pm

Re: Backtesting some robo-advisers asset allocations

Post by billthecat » Wed Aug 02, 2017 12:20 am

siamond wrote: Well, you will find that quite a few people on this board would challenge this 'conventional wisdom', and would much prefer staying with a fixed asset allocation for good.
OK. That's interesting. Didn't Jack Bogle himself say age in bonds? Not that you have to follow Bogle to be on bogleheads.org.

User avatar
in_reality
Posts: 4219
Joined: Fri Jul 12, 2013 6:13 am

Re: Backtesting some robo-advisers asset allocations

Post by in_reality » Wed Aug 02, 2017 1:39 am

siamond wrote: Ok, so I guess you're a smart beta believer. Fair enough, I respect that. Please appreciate that eight years of over performance doesn't demonstrate much though (heck, small value underperformed TSM for *decades* in the past). I am not saying you're wrong, mind you, I am just saying we don't have the proper historical data to assess the past on a significant enough time period (i.e. several decades), nor to backtest. Plus new strategies have this tendency of doing well to begin with, then falter. Future will tell. Your point is well taken, though.
No, I'm not a smart beta believer. Some if it is over valued as a result of popularity and recent performance attracting attention.

I use the Fundamental Index for their value exposure though - International small cap has done quite well. I overweight that much more than SIP does.

I agree the time is too short to draw any conclusion ... and also question people's reducing their tilt to simplify things when they get older.

So about SIP, how will it do in a rising rate environment? Sure a high allocation to TBM did well in a great bond bull market

Also, do tilters really maintain their tilts or abandon them for simplification as time goes on. I plan on maintaining a static allocation but wonder if my wife would do the same.

Anyway, thanks for your analysis. Great sheets!
[60% US _ 26% DEV _ 14% EM] | (-16% LC _ +8% MC _ +8% SC) | [47% FND/VAL _ 40% MKT _ 7% MOM _ 6% REIT] | (+/- 5% or *25% rebalancing bands)

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 5:10 am

siamond wrote:
jbolden1517 wrote:As far as the comparison with 3-fund. IMHO Tilt1 and Tilt2 IMHO are nothing like 3 fund, those are more than tilts they are entirely different asset allocation. I think Tilt2 demonstrates that the more you move away from large cap growth and into value the better.
I don't quite agree. To build Tilt1, I just took some % out of TSM and sprinkled it to overload SCV and REITs (and same for Int'l). This is truly how Bogleheads tilters do it, as far as I understand. Tilt2 is more aggressive for sure, and less people would go there, although the historical evidence is indeed very strong. Now will the future rhyme with the past, we shall see...
My claim is that 3-fund is a bad asset allocation because of large growth. When you make it no longer large growth you don't have that problem. I agree that Tilt1 and Tilt2 are the sort of thing Bogleheads do.
siamond wrote:
jbolden1517 wrote:The one thing that is interesting about SIP is how lousy the fund was 1984-1990. That's the period that did the damage, after that it keeps up with Tilt2 (the clear winner). I would have thought the high yield in the mid 1980s would have been a huge performance kick. Could you let me know what's happening in that period that's causing the underperformance?
The drag from the 9% cash is is the main issue, just moving it to bonds basically fixes the portfolio and brings it back to par with my tilting portfolios. Precious metals also hurt a bit. I suppose SIP uses a fairly extreme rebalancing technique requiring such amount of cash (as opposed to the simple annual rebalancing I assumed), but I've run enough simulations of rebalancing algorithms to be very skeptical about their ability to generate a true premium.
True, there is a momentum effect the faster you rebalance the more that hurts. I think they do rebalance quite often especially early on for TLH. On the website they indicated that clients shouldn't expect a need for daily rebalancing. So I'm going to assume something like weekly.

siamond wrote: IMPORTANT EDIT: just realized that I was too literal in my interpretation of cash. I interpreted cash as hard cash, something returning 0% a year. But coming back to Schwab's guide of the funds they use, of course, they do the obvious and use a money-market fund. So I should have mapped the 9% cash to something like VUSXX in my backtesting. I just gave it a quick try, and this definitely helps make SIP better. Still not as good as simply put those 9% in Total-Bonds, but this does help.
Ah yeah that will matter especially in those early years when cash was returning well. Not so much since 2008. Can you throw the Bogle graph back?
Last edited by jbolden1517 on Wed Aug 02, 2017 5:32 am, edited 1 time in total.

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 5:14 am

siamond wrote:This being said, going to 60/40 or 50/50 is one thing (and that's perfectly reasonable for many people), while going to 30/70 is another (that is quite extreme, and can really backfire after a couple of decades of retirement - this is what I was alluding to).
We are in perfect agreement there. People are forgetting that bond bears in the USA are a generation long problem not something that last a few years. Inflation hits retirement investors need to draw at least at the CPI (some argue it underestimates heavily).

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 5:24 am

siamond wrote:[quote="in_reality"
in_reality wrote:Also, why is it important to break US stocks down into small value, but not important for international? OK lack of data. But SIP does give small value international developed exposure. It also gives emerging value exposure (large and mid-cap too).
And there is no such historical data that I am aware of. I made clear in this post that I had to make a few choices like that. I seriously doubt this changes the big picture though.
I wouldn't be so sure about that it doesn't change things it is an 18% extra value tilt. That should be at least another 50 basis points (and I suspect with EMs a lot more since those are way more bubbly). Also just to mention again across the board you have the issue that SIP is deeper value than Vanguard, their fundamentals tilt harder. What about using something like an active value fund (say Templeton) and adding all but 50 basis points of ER back in for the back test?

longinvest
Posts: 2467
Joined: Sat Aug 11, 2012 8:44 am

Re: Backtesting some robo-advisers asset allocations

Post by longinvest » Wed Aug 02, 2017 6:29 am

Jbolden1517,
jbolden1517 wrote: My claim is that 3-fund is a bad asset allocation because of large growth.
Let's consider your statement as a universal truth. Let's see what mathematical reasoning tells us about it.

Hypothesis: An asset allocation containing large growth is always a bad asset allocation.

Consequence: Investors, being rational, avoid this asset subclass in their asset allocation. In doing so, the price of large growth drops. At some point, this subclass will have the best future returns; it cannot be otherwise. At that point, the hypothesis becomes false, leading to a contradiction.

Therefore, your statement is not a universal truth.

In other words, your statement is false.

I love mathematics.
Bogleheads investment philosophy | Lifelong Portfolio: 25% each of (domestic/international)stocks/(nominal/inflation-indexed)bonds | VCN/VXC/VAB/ZRR

User avatar
in_reality
Posts: 4219
Joined: Fri Jul 12, 2013 6:13 am

Re: Backtesting some robo-advisers asset allocations

Post by in_reality » Wed Aug 02, 2017 6:34 am

jbolden1517 wrote:
siamond wrote:[quote="in_reality"
in_reality wrote:Also, why is it important to break US stocks down into small value, but not important for international? OK lack of data. But SIP does give small value international developed exposure. It also gives emerging value exposure (large and mid-cap too).
And there is no such historical data that I am aware of. I made clear in this post that I had to make a few choices like that. I seriously doubt this changes the big picture though.
I wouldn't be so sure about that it doesn't change things it is an 18% extra value tilt. That should be at least another 50 basis points (and I suspect with EMs a lot more since those are way more bubbly). Also just to mention again across the board you have the issue that SIP is deeper value than Vanguard, their fundamentals tilt harder. What about using something like an active value fund (say Templeton) and adding all but 50 basis points of ER back in for the back test?
Additional things working against the SIP returns are:

1) the SIP emerging bonds got allocated to International bonds. I suspect the international bond fund is hedge and so would have a nice return from dropping international rates in USD terms. The USD has appreciated 30% in recent years and so I think emerging bonds in local currencies haven't returned anything. Long term you are rebalancing into them and I suspect they will do well when the USD falls. If it doesn't, economic competitiveness suggests that foreign companies can make inroads to the US with cheaper products and also expand market share at home.

2) the International REITS have been allocated to REITs I think which are US. International hasn't done nearly as well.

I mean honestly, I expect SIP to underperform somewhat. Inflation protections has a cost and any portfolio being heavier in bonds during the bull will surely do better.
[60% US _ 26% DEV _ 14% EM] | (-16% LC _ +8% MC _ +8% SC) | [47% FND/VAL _ 40% MKT _ 7% MOM _ 6% REIT] | (+/- 5% or *25% rebalancing bands)

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 6:40 am

longinvest wrote: Let's consider your statement as a universal truth. Let's see what mathematical reasoning tells us about it.

Hypothesis: An asset allocation containing large growth is always a bad asset allocation.
I didn't say that. An asset allocation using large growth as a core is always a bad asset allocation.
longinvest wrote: Consequence: Investors, being rational, avoid this asset subclass in their asset allocation.
You are forgetting 4 important features that cause growth to underperform even with rational investors. The most important of which is that the investor in this case is being perfectly rational. Investment managers mutual fund managers are compensated for assets under management not return. Pension fund managers are compensated usually on 3 year track records relative to a benchmark. Their payoff matrix is not the same as the investor's payoff matrix. The managers not shockingly follow strategies designed to grow their payoff even at the expense of return. So for example: growth stocks have many more highly positive 3 year periods than value stocks (which effects both mutual fund and pension fund managers, and in today's world likely hedge fund managers). Value stocks outperform during periods of time when new investment in mutual funds are low, growth stocks outperform at times when they are high....

Another one that's important is the winners curse. Assume there is an auction with S sellers and B buyers each of which is capable of evaluating an item and making a buy sell decision based on the price relative to their evaluation. Assume each of their individual evaluations has random error. if S > B the asset will tend to sell for below fair price while if B > S the assets will tend to sell above fair price. This becomes exploitable if one merely makes blind buy/sell decisions based on the ratio between S and B.

I'll leave it at those 2 for now since they capture the flavor of where your thinking is flawed.
longinvest wrote: In doing so, the price of large growth drops. At some point, this subclass will have the best future returns; it cannot be otherwise.
That does happen to the individual stocks. It can't happen to the subclass, because the subclass isn't static with respect to what stocks it holds. When the stocks drop they become value stocks and do have the best future returns.
longinvest wrote: At that point, the hypothesis becomes false, leading to a contradiction. Therefore, your statement is not a universal truth. In other words, your statement is false. I love mathematics.
That's not mathematics. Its religion. Mathematics is more careful about the types of assumptions you were making.

longinvest
Posts: 2467
Joined: Sat Aug 11, 2012 8:44 am

Re: Backtesting some robo-advisers asset allocations

Post by longinvest » Wed Aug 02, 2017 6:43 am

Everytime I read false statements, I always ask myself: why would someone tell me such lies?

In investing, we should always remember that there are two parties in every transaction: the seller and the buyer.

Someone wanting me to sell a security at a lower price than intrinsic value will try to convince me and lots of other people that I should not keep this security in my portfolio. If believe him and act on it, he'll gladly be on the other side of my transaction in full anonymity to buy it from me at a low price.

Someone wanting me to buy a security at a higher price than intrinsic value will try to convince me and lots of other people that I should add more of this security into my portfolio. If believe him and act on it, he'll gladly be on the other side of my transaction in full anonymity to sell it to me at a high price.

Bogleheads, beware of anybody claiming that the three-fund portfolio (total-market domestic stocks, total-market domestic stocks, total-market domestic nominal bonds), or the four-fund portfolio (total-market domestic stocks, total-market domestic stocks, total-market domestic nominal bonds, total-market domestic inflation-indexed bonds) are bad portfolios.
Bogleheads investment philosophy | Lifelong Portfolio: 25% each of (domestic/international)stocks/(nominal/inflation-indexed)bonds | VCN/VXC/VAB/ZRR

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 6:47 am

in_reality wrote:
jbolden1517 wrote:
siamond wrote:[quote="in_reality"
in_reality wrote:Also, why is it important to break US stocks down into small value, but not important for international? OK lack of data. But SIP does give small value international developed exposure. It also gives emerging value exposure (large and mid-cap too).
And there is no such historical data that I am aware of. I made clear in this post that I had to make a few choices like that. I seriously doubt this changes the big picture though.
I wouldn't be so sure about that it doesn't change things it is an 18% extra value tilt. That should be at least another 50 basis points (and I suspect with EMs a lot more since those are way more bubbly). Also just to mention again across the board you have the issue that SIP is deeper value than Vanguard, their fundamentals tilt harder. What about using something like an active value fund (say Templeton) and adding all but 50 basis points of ER back in for the back test?
Additional things working against the SIP returns are:
in_reality wrote: 1) the SIP emerging bonds got allocated to International bonds. I suspect the international bond fund is hedge and so would have a nice return from dropping international rates in USD terms. The USD has appreciated 30% in recent years and so I think emerging bonds in local currencies haven't returned anything. Long term you are rebalancing into them and I suspect they will do well when the USD falls. If it doesn't, economic competitiveness suggests that foreign companies can make inroads to the US with cheaper products and also expand market share at home.
You are correct. That's not even a close match. EM bonds have a very high return.
in_reality wrote: 2) the International REITS have been allocated to REITs I think which are US. International hasn't done nearly as well.
Are you sure international REITs haven't done well. I do know of active funds that would work for a proxy. Outside of Europe real estate development has been hugely profitable and in Europe you have had very low interest rates (subsidized) on real estate loans for a long time.

As for your comment on inflation protection, agree. If we included the early 1970s this portfolio does better. Since we are somewhere in the late 1930s right now in terms of economic cycle (without WWII likely to intervene) things should play out quite differently than the bond bull. But the experiment was to backtest the way Bogleheads backtest portfolios. I think the problem with Simba is that as we start moving away from Vanguard style allocations it doesn't work well. Betterment portfolios would backtest much more easily than SIP. And as we discussed in the thread, Hedgeable isn't testable at all since it doesn't maintain fixed allocations.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Wed Aug 02, 2017 6:52 am

jbolden1517 wrote:Ah yeah that will matter especially in those early years when cash was returning well. Not so much since 2008. Can you throw the Bogle graph back?
I will redo all the graphs later today, the cash vs. money market mistake was too glaring, I really need to fix that.

PS. please review carefully the red mappings in this post. If you have a better idea than what I did, using the *existing* asset classes we have in Simba, speak up, I'm happy to change those too. I just don't want to introduce a new improvised data series in there - at least for now.

PS2. Int'l REITs didn't go well so far, but very little history is available for it... This year is better, actually!
Last edited by siamond on Wed Aug 02, 2017 6:59 am, edited 1 time in total.

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 6:59 am

in_reality wrote: 1) the SIP emerging bonds got allocated to International bonds. I suspect the international bond fund is hedge and so would have a nice return from dropping international rates in USD terms. The USD has appreciated 30% in recent years and so I think emerging bonds in local currencies haven't returned anything. Long term you are rebalancing into them and I suspect they will do well when the USD falls. If it doesn't, economic competitiveness suggests that foreign companies can make inroads to the US with cheaper products and also expand market share at home.
You are correct. That's not even a close match. EM bonds have a very high return.
in_reality wrote: 2) the International REITS have been allocated to REITs I think which are US. International hasn't done nearly as well.
Are you sure international REITs haven't done well. I do know of active funds that would work for a proxy. Outside of Europe real estate development has been hugely profitable and in Europe you have had very low interest rates (subsidized) on real estate loans for a long time.

As for your comment on inflation protection, agree. If we included the early 1970s this portfolio does better. Since we are somewhere in the late 1930s right now in terms of economic cycle (without WWII likely to intervene) things should play out quite differently than the bond bull. But the experiment was to backtest the way Bogleheads backtest portfolios. I think the problem with Simba is that as we start moving away from Vanguard style allocations it doesn't work well. Betterment portfolios would backtest much more easily than SIP. And as we discussed in the thread, Hedgeable isn't testable at all since it doesn't maintain fixed allocations.[/quote]

User avatar
in_reality
Posts: 4219
Joined: Fri Jul 12, 2013 6:13 am

Re: Backtesting some robo-advisers asset allocations

Post by in_reality » Wed Aug 02, 2017 7:18 am

jbolden1517 wrote:
in_reality wrote: 2) the International REITS have been allocated to REITs I think which are US. International hasn't done nearly as well.
Are you sure international REITs haven't done well. I do know of active funds that would work for a proxy. Outside of Europe real estate development has been hugely profitable and in Europe you have had very low interest rates (subsidized) on real estate loans for a long time.

Since 12.2010 US REITs have had a 2.6% higher CAGR. That's the start of Vanguards international version. Again, the USD has boomed so even if returns were ok in local terms, it's fighting an uphill battle for the time-being.
[60% US _ 26% DEV _ 14% EM] | (-16% LC _ +8% MC _ +8% SC) | [47% FND/VAL _ 40% MKT _ 7% MOM _ 6% REIT] | (+/- 5% or *25% rebalancing bands)

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 7:57 am

in_reality wrote:
jbolden1517 wrote:
in_reality wrote: 2) the International REITS have been allocated to REITs I think which are US. International hasn't done nearly as well.
Are you sure international REITs haven't done well. I do know of active funds that would work for a proxy. Outside of Europe real estate development has been hugely profitable and in Europe you have had very low interest rates (subsidized) on real estate loans for a long time.
Since 12.2010 US REITs have had a 2.6% higher CAGR. That's the start of Vanguards international version. Again, the USD has boomed so even if returns were ok in local terms, it's fighting an uphill battle for the time-being.
Ah I meant longer term where the currency fluctuations wouldn't matter so much. I will say this using Vanguard as a proxy rather than market indexes or more similar funds is definitely creating problems in back testing methodology.

Ethelred
Posts: 211
Joined: Sun Oct 30, 2016 9:38 am

Re: Backtesting some robo-advisers asset allocations

Post by Ethelred » Wed Aug 02, 2017 8:38 am

This is interesting, and great analysis, thanks Siamond.
siamond wrote:I don't quite agree. To build Tilt1, I just took some % out of TSM and sprinkled it to overload SCV and REITs (and same for Int'l). This is truly how Bogleheads tilters do it, as far as I understand. Tilt2 is more aggressive for sure, and less people would go there, although the historical evidence is indeed very strong. Now will the future rhyme with the past, we shall see...
That sounds reasonable, though I'm not sure how that REITs are that popular for tilting these days. The SCV funds are used to capture the small and value factors, and some Bogleheads will also tilt to momentum and other factors, though these are more expensive to access. However, real estate is not itself a factor.

I'm not sure how much of an impact you'd get from moving 10% REITs to SCV in portfolio 5 (or 5% to SCV and 5% to momentum), but it could be significant.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Wed Aug 02, 2017 9:18 am

Ethelred wrote:That sounds reasonable, though I'm not sure how that REITs are that popular for tilting these days. The SCV funds are used to capture the small and value factors, and some Bogleheads will also tilt to momentum and other factors, though these are more expensive to access. However, real estate is not itself a factor.

I'm not sure how much of an impact you'd get from moving 10% REITs to SCV in portfolio 5 (or 5% to SCV and 5% to momentum), but it could be significant.
Based on the threads I've read in the past few years, REITs remain a fairly popular tilt among a good number of Bogleheads. But sure, there are many individual variations in one's choice for tilts. As to replacing the REIT tilt by more SCV, this does change the (past) trajectory a tad, but doesn't create such a big difference, which is a well-known thing, REITs and SCV do overlap a significant manner (or at least did in the past).

Ethelred
Posts: 211
Joined: Sun Oct 30, 2016 9:38 am

Re: Backtesting some robo-advisers asset allocations

Post by Ethelred » Wed Aug 02, 2017 11:43 am

longinvest wrote:Let's consider your statement as a universal truth. Let's see what mathematical reasoning tells us about it.

Hypothesis: An asset allocation containing large growth is always a bad asset allocation.

Consequence: Investors, being rational, avoid this asset subclass in their asset allocation. In doing so, the price of large growth drops. At some point, this subclass will have the best future returns; it cannot be otherwise. At that point, the hypothesis becomes false, leading to a contradiction.

Therefore, your statement is not a universal truth.

In other words, your statement is false.

I love mathematics.
I'm sorry, that's not mathematics, it's faulty logic based on an unproven assumption, and presented using remarkably dogmatic language.

If you can prove that all investors act rationally, your argument could be correct. However, it's easy to reverse your argument to prove that investors are not rational, at least in your terms: Whenever any asset class drops in price relative to others, rational investors will buy those assets. As a direct result, there should be no difference in return between US and international stocks, small or large stocks, value or growth stocks, and there should be no stock market crashes because rational investors would immediately buy up the undervalued stocks, so restoring market equilibrium.

Ethelred
Posts: 211
Joined: Sun Oct 30, 2016 9:38 am

Re: Backtesting some robo-advisers asset allocations

Post by Ethelred » Wed Aug 02, 2017 11:43 am

siamond wrote:
Ethelred wrote:That sounds reasonable, though I'm not sure how that REITs are that popular for tilting these days. The SCV funds are used to capture the small and value factors, and some Bogleheads will also tilt to momentum and other factors, though these are more expensive to access. However, real estate is not itself a factor.

I'm not sure how much of an impact you'd get from moving 10% REITs to SCV in portfolio 5 (or 5% to SCV and 5% to momentum), but it could be significant.
Based on the threads I've read in the past few years, REITs remain a fairly popular tilt among a good number of Bogleheads. But sure, there are many individual variations in one's choice for tilts. As to replacing the REIT tilt by more SCV, this does change the (past) trajectory a tad, but doesn't create such a big difference, which is a well-known thing, REITs and SCV do overlap a significant manner (or at least did in the past).
Thanks. I did wonder if that might be the case.

garlandwhizzer
Posts: 1607
Joined: Fri Aug 06, 2010 3:42 pm

Re: Backtesting some robo-advisers asset allocations

Post by garlandwhizzer » Wed Aug 02, 2017 2:40 pm

Wow! What an extraordinary piece of work and research, siamond. Thanks for sharing it.

I have so many thoughts about the SIP as well as the 3 fund and tilted approaches. It seems to me that each one has a rational basis but that does not guarantee outperformance. I am not sure what conclusion to draw from all this data about the optimal approach starting from where we are today for the long term future.

First, it seems to me that all portfolios that were tested had good and acceptable results. The CAGR differences between them was less than 1%, between the best of the five and the worst of the five, and even the worst (70/30 base) managed 9.2% over the long time period. I would be overjoyed if my portfolio returns that over a long time frame.

Second, the question arises: will these results over a very long time period in the past be projected into future results going forward for a long time period? The case can be made that markets are more professionally dominated now than in the distant past, presumably more efficient, and that factor investing has become popular and perhaps an overcrowded trade relative to the distant past when such approaches weren't common knowledge. It is likely IMO that factor approaches will yield less impressive results going forward than in the long term past.

Third, the case can also be made that our economy is now structural different than in much of the historical past--higher levels of personal, corporate, and governmental debt, aging demographics, longer lifespans with relatively more years of retirement, consuming assets, relative to years of employment, producing assets, declining productivity growth, and finally the "new normal"--secular slow economic growth, low inflation, low bond yields unresponsive to maximal monetary stimulus for 9 years. In much of our economic past, especially the 1980s and 1990s, we had robust growth pumping up corporate profits. Also stock prices have during the 34 year bond bull market that started in 1982 benefited from the constant tailwinds of decreasing bond yields and interest rates. PE raitios of the S&P500 have increased from less than 8 in the early 1980s to about 25 now. Those wonderful years appear to be behind us now and what worked then may not work as well going forward.

In other words, a lot has changed over the past 50+ years. Does this affect the reliability of data mining as a predictor of future returns? I don't know the answer to that question but I am a skeptical person by nature. I don't think it is predictable accurately now, that it is more a question of faith than science. It seems to me that the good news is that you don't have to choose the perfect portfolio beforehand to obtain good results. Bet on where your heart and mind are in terms of these 5 choices, and if you hold on for long enough, you are likely to do very well. Finding a good solid portfolio for the long run is not difficult. The hard part is staying the course. The impossible part is picking ex ante the perfect portfolio for your specific time span in the future, although it certainly is tempting to try.

Garland Whizzer

longinvest
Posts: 2467
Joined: Sat Aug 11, 2012 8:44 am

Re: Backtesting some robo-advisers asset allocations

Post by longinvest » Wed Aug 02, 2017 2:59 pm

Ethelred,
jbolden1517 wrote: My claim is that 3-fund is a bad asset allocation because of large growth.
Ethelred wrote:
longinvest wrote:Let's consider your statement as a universal truth. Let's see what mathematical reasoning tells us about it.

Hypothesis: An asset allocation containing large growth is always a bad asset allocation.

Consequence: Investors, being rational, avoid this asset subclass in their asset allocation. In doing so, the price of large growth drops. At some point, this subclass will have the best future returns; it cannot be otherwise. At that point, the hypothesis becomes false, leading to a contradiction.

Therefore, your statement is not a universal truth.

In other words, your statement is false.

I love mathematics.
I'm sorry, that's not mathematics, it's faulty logic based on an unproven assumption, and presented using remarkably dogmatic language.

If you can prove that all investors act rationally, your argument could be correct. However, it's easy to reverse your argument to prove that investors are not rational, at least in your terms: Whenever any asset class drops in price relative to others, rational investors will buy those assets. As a direct result, there should be no difference in return between US and international stocks, small or large stocks, value or growth stocks, and there should be no stock market crashes because rational investors would immediately buy up the undervalued stocks, so restoring market equilibrium.
I've used a traditional technique to prove something false: (1) state a hypothesis, (2) assume that it is true, and (3) show that this leads to a contradiction.

I've stated a hypothesis. Assuming it is true, why would any investor go and invest into an ever losing asset class? Not all investors need to be rational, just enough of them.

Now, if you have trouble with this technique, let me state things differently.

There are many ways to prove that jbolden1517's statement is false. A single counterexample is sufficient. That's how logical reasoning works. It is actually very easy to find a past period of time when an asset allocation containing large growth was a good allocation. You know as well as I do that large growth has outperformed other asset classes such as small-cap value at some point in the past. It doesn't have to be always; just once is sufficient to use as a counterexample to disprove jbolden1517's statement.

It is up to the person claiming that a three-fund portfolio is a bad portfolio to make a mathematical proof that it has always, and will always be so. Unfortunately, such a proof is impossible to make, because I have proved the claim to be false.

I didn't choose to disprove the statement using a simple counterexample, in my quoted post, because I wanted to insist on the more fundamental concept of future returns going up when investors flee an asset class.

More interestingly, let's assume that someone discovered a really bullet-proof approach to beat the market. Why would that person go and reveal it using an anonymous nickname on a public investing forum, instead of using it to become the next Warren Buffet? Wouldn't that be counter productive, as we know that due to the law of supply and demand, if more people know about it, the approach will lose its advantage over the market?

Total-market indexing is a good investing approach. It is simple and guarantees that the investor will harvest the returns of the market (minus a ridiculously small fee, when using low-cost funds and ETFs). Even if everybody adopted index investing, everybody would simply harvest market returns (minus fees). This universality is a special property that is unique to total-market indexing. So, one should beware of anybody claiming that total-market investing is a bad approach to investing.
Bogleheads investment philosophy | Lifelong Portfolio: 25% each of (domestic/international)stocks/(nominal/inflation-indexed)bonds | VCN/VXC/VAB/ZRR

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Wed Aug 02, 2017 3:22 pm

I will update all the charts tonight, to correct my mistake about the cash allocation, which means 'money market' and not cold hard cash. This essentially makes SIP on par with the Tilt1 portfolio. Which admittedly makes more sense.

If anybody has another practical suggestion to improve the approximate mapping described in this post, please speak up...

Note that for the 7% emerging bonds, I had mapped half of them to Int'l bonds, and half of them to HY bonds. This was my weak attempt to not undermine the potential return, risk and low correlation of EM bonds. Admittedly very approximate, but we seriously lack historical data here (including periods of time with defaults).
Last edited by siamond on Wed Aug 02, 2017 3:32 pm, edited 1 time in total.

LocusCoeruleus
Posts: 4
Joined: Wed May 24, 2017 10:00 pm

Re: Backtesting some robo-advisers asset allocations

Post by LocusCoeruleus » Wed Aug 02, 2017 3:28 pm

Siamond- a request to do similar analysis for wealthfront & betterment. I love that telltale chart. Thanks!

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Wed Aug 02, 2017 3:35 pm

LocusCoeruleus wrote:Siamond- a request to do similar analysis for wealthfront & betterment. I love that telltale chart. Thanks!
Yes, I plan to do that. Just want to make sure SIP is fully sorted out and that I got most kinks out of the process before I get to new AAs. Feel free to download the spreadsheet I posted here to experiment by yourself.

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 6:44 pm

siamond wrote:I will update all the charts tonight, to correct my mistake about the cash allocation, which means 'money market' and not cold hard cash. This essentially makes SIP on par with the Tilt1 portfolio. Which admittedly makes more sense.

If anybody has another practical suggestion to improve the approximate mapping described in this post, please speak up...

Note that for the 7% emerging bonds, I had mapped half of them to Int'l bonds, and half of them to HY bonds. This was my weak attempt to not undermine the potential return, risk and low correlation of EM bonds. Admittedly very approximate, but we seriously lack historical data here (including periods of time with defaults).
That's going to be too low a correlation. What about 50% EM stock, 50% INTL bonds. I don't love it but unless you have intl money markets (which vanguard doesn't carry for USA investors) I don't see how to do something like INTL bonds on leverage (which might be a better approximation).

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Wed Aug 02, 2017 7:07 pm

jbolden1517 wrote:
siamond wrote:I will update all the charts tonight, to correct my mistake about the cash allocation, which means 'money market' and not cold hard cash. This essentially makes SIP on par with the Tilt1 portfolio. Which admittedly makes more sense.

If anybody has another practical suggestion to improve the approximate mapping described in this post, please speak up...

Note that for the 7% emerging bonds, I had mapped half of them to Int'l bonds, and half of them to HY bonds. This was my weak attempt to not undermine the potential return, risk and low correlation of EM bonds. Admittedly very approximate, but we seriously lack historical data here (including periods of time with defaults).
That's going to be too low a correlation. What about 50% EM stock, 50% INTL bonds. I don't love it but unless you have intl money markets (which vanguard doesn't carry for USA investors) I don't see how to do something like INTL bonds on leverage (which might be a better approximation).
Yes, I am having qualms as well, and this 7% position is quite significant. I started to research what available data we do have (e.g. Barclays indices and so on), I know this doesn't go very far in time, but this should allow to do a basic sanity-check of whatever coarse rule we might use. Your idea is interesting. About MBS, we might actually be able to build a decent data series, here indices do go back farther in time. This will take me a bit longer though, but those are fair questions, and might lead to possible Simba improvements.

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 7:11 pm

longinvest wrote: I've used a traditional technique to prove something false: (1) state a hypothesis, (2) assume that it is true, and (3) show that this leads to a contradiction.
Except you
1) Mischaracterized the hypothesis
2) Didn't really assume it was true
3) Didn't show a contradiction.
longinvest wrote: I've stated a hypothesis. Assuming it is true, why would any investor go and invest into an ever losing asset class? Not all investors need to be rational, just enough of them.
I gave you two reasons. There are many more. Address those. BTW no one is claiming that large growth is an ever losing asset class. Large growth outperforms value in most 3 year periods was clearly stated by me. Where it fails is 20 year periods.
longinvest wrote: It is up to the person claiming that a three-fund portfolio is a bad portfolio to make a mathematical proof that it has always, and will always be so.
I've never said it always has and always will be so. It is a bad portfolio now. It is a bad portfolio in most situations. It would be an excellent portfolio in a period of time when large growth is underpriced (say investors are systematically underestimating earnings growth) and the efficiencies provided by indexing and cap weighting outweigh the costs (say for example a very high capital gains tax rate).
longinvest wrote: More interestingly, let's assume that someone discovered a really bullet-proof approach to beat the market. Why would that person go and reveal it using an anonymous nickname on a public investing forum, instead of using it to become the next Warren Buffet? Wouldn't that be counter productive, as we know that due to the law of supply and demand, if more people know about it, the approach will lose its advantage over the market?
I'm just a nice guy I guess. And besides even if I did have that kind of influence to move trillions of dollars, I have to live in this country not just invest in it. Reckless investors buying bad assets and thus distorting the economy and market does me lots of harm.
longinvest wrote: Total-market indexing is a good investing approach. It is simple and guarantees that the investor will harvest the returns of the market (minus a ridiculously small fee, when using low-cost funds and ETFs).
No it doesn't. Indexers are just another group of traders. You don't sit above the market looking on from afar, you participate in it. When index funds buy using their float based trading strategy they distort market prices to reward dilution and punish concentration. That changes market behavior. Which wouldn't matter too much. Except that indexing and quasi indexing is now 70% of the SP500. So their trading strategy changes economic behavior and that does matter to every American myself included.
longinvest wrote: Even if everybody adopted index investing, everybody would simply harvest market returns (minus fees).
Which would then be likely be highly negative returns because at that point any company could simply dilute and inject capital into its business regardless of how efficiently it was using capital. It could through synthetic dilution raise its share price infinitely. And of course through even minute concentration it could induce an infinite cascade of sell orders allowing any control investor to take any successful business private at infinitely low cost.

You aren't thinking. Yes they would capture the "market return" which would be rather quickly -100%.
longinvest wrote: This universality is a special property that is unique to total-market indexing.
Nonsense. If there were a large group of investors who were sales weighting (like the Schwab fundamental funds we are supposed to be talking about ) stock prices would align so that P/S ratios were normalized across the market. Everyone could hold a equal percentage of corporate sales. If the market were dividend weighted, P/D ratios would normalize across the market... float*shareprice (market cap) is just another number. A float based trading strategy is just another trading strategy. It happens to be popular, and it is very inexpensive to administrate. That's it There is no magic.

[OT comment removed by moderator prudent]

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Wed Aug 02, 2017 7:44 pm

siamond wrote:
jbolden1517 wrote:
siamond wrote:I will update all the charts tonight, to correct my mistake about the cash allocation, which means 'money market' and not cold hard cash. This essentially makes SIP on par with the Tilt1 portfolio. Which admittedly makes more sense.

If anybody has another practical suggestion to improve the approximate mapping described in this post, please speak up...

Note that for the 7% emerging bonds, I had mapped half of them to Int'l bonds, and half of them to HY bonds. This was my weak attempt to not undermine the potential return, risk and low correlation of EM bonds. Admittedly very approximate, but we seriously lack historical data here (including periods of time with defaults).
That's going to be too low a correlation. What about 50% EM stock, 50% INTL bonds. I don't love it but unless you have intl money markets (which vanguard doesn't carry for USA investors) I don't see how to do something like INTL bonds on leverage (which might be a better approximation).
Yes, I am having qualms as well, and this 7% position is quite significant. I started to research what available data we do have (e.g. Barclays indices and so on), I know this doesn't go very far in time, but this should allow to do a basic sanity-check of whatever coarse rule we might use. Your idea is interesting. About MBS, we might actually be able to build a decent data series, here indices do go back farther in time. This will take me a bit longer though, but those are fair questions, and might lead to possible Simba improvements.
I do have data for the last 20 years. The green box JPM-EMBI is the index that Schwab's fund (VanEck) is tracking.
http://www.lazardnet.com/us/docs/sp0/20 ... nvestmentF

The construction of the index is essentially: countries that aren't painful to deal (either legally or practically) with and have bond market liquid enough to stick lots of money in. So I'd assume the returns will fairly closely match the index.

Those other indexes might be nice for the sheet as well.

That being said this data doesn't go back far enough. But at least it might allow for a good model of what the index looks like.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Wed Aug 02, 2017 8:00 pm

*** IMPORTANT UPDATE ***

I updated the graphs and the commentary of the initial posts while re-allocating cash to a money market fund (VUSXX in Simba). This makes SIP (more precisely, its Simba approximation) look better, although still not quite convincing imho.


I didn't touch the other mappings for now. Let's take more time to think about new data series to try to better capture EM bonds, MBS, etc. Thank you for all the feedback, this is making progress and is quite interesting.

PS. if we could keep this thread focused on its primary topic, this would be appreciated. Discussions about the validity of the value factor theory, or 'fundamental' vs. 'classic' value, would be better addressed with their own threads (and there are plenty of those already!)... Pretty please?

User avatar
LadyGeek
Site Admin
Posts: 40469
Joined: Sat Dec 20, 2008 5:34 pm
Location: Philadelphia
Contact:

Re: Backtesting some robo-advisers asset allocations

Post by LadyGeek » Wed Aug 02, 2017 8:01 pm

I think this discussion provides a necessary perspective in that robo-advisers are just that - advisers. Portfolio performance is no different for a robo-adviser than anyone else.

I've added a new section in the wiki to address this. See: Robo-adviser (Portfolio performance)

Comments / questions / concerns are welcome.
To some, the glass is half full. To others, the glass is half empty. To an engineer, it's twice the size it needs to be.

User avatar
LadyGeek
Site Admin
Posts: 40469
Joined: Sat Dec 20, 2008 5:34 pm
Location: Philadelphia
Contact:

Re: Backtesting some robo-advisers asset allocations

Post by LadyGeek » Wed Aug 02, 2017 8:32 pm

LocusCoeruleus wrote:Siamond- a request to do similar analysis for wealthfront & betterment. I love that telltale chart. Thanks!
I also like that telltale chart, which deserves more explanation than what's in the backtesting spreadsheet.

Next up for the wiki, a deep-dive into the details of the telltale chart. siamond has given me some pointers.
To some, the glass is half full. To others, the glass is half empty. To an engineer, it's twice the size it needs to be.

User avatar
siamond
Posts: 3328
Joined: Mon May 28, 2012 5:50 am

Re: Backtesting some robo-advisers asset allocations

Post by siamond » Thu Aug 03, 2017 8:35 am

jbolden1517 wrote:That's going to be too low a correlation. What about 50% EM stock, 50% INTL bonds.
[...]
I do have data for the last 20 years. The green box JPM-EMBI is the index that Schwab's fund (VanEck) is tracking.
http://www.lazardnet.com/us/docs/sp0/20 ... nvestmentF
Thank you, great pointer, much better than Barclays or Citi. JPM-EMBI matches reasonably well the (very few!) years of the corresponding Vanguard fund (VGOVX), and this index does provide pretty good EM Bonds history, starting from 1994.

As to the previous years, trying to fill the gap 1985 to 1993, the best I found is 1/3rd EM, 1/3rd HY, 1/3rd Int'l Bonds. Yes, VERY clunky, but definitely a better match for std-deviation and correlation with JPM EMBI for known years than other things I tried.
=> side note: this made me understand why SIP (and other folks) are eager to use EM Bonds, as this asset class has an interestingly low correlation (e.g. less than 0.5) with pretty much everything else. I learn something every day with this forum!

So... I cobbled together an extended JPM EMBI data series starting from 1985. And assigned the 7% EM Bonds to it. And lo and behold, the SIP results are indeed improved, and now it's roughly on par with my Tilt2 portfolio. And I guess that if I were to address the more minor issues (e.g. better MBS mapping, etc), this would improve a tad further. Then the big thing remains 'fundamental' vs. 'classic' value, but let's stop short of re-opening this discussion (please!). Note that one could of course also use 'fundamental' funds in a regular tilted asset allocation (some Bogleheads definitely do that).

I also tried to replace the 9% money market 'cash' by TBM, which makes the whole thing even better - just saying! This is the part of the SIP portfolio I really don't understand. Here is the resulting telltale (SIP Base is my previous definition of SIP; SIP EMB uses the JPM EM index; SIP TBM uses the JPM EM index and replaces cash by TBM):

Image

Bottomline: SIP is basically a somewhat complicated value-y asset allocation, with a few interesting subtleties in terms of diversification and play on correlations. Although one can construct a simpler tilted portfolio with regular Bogleheads principles that would have performed more or less the same. As Ladygeek said, the Schwab robo-adviser essentially acted as an adviser with its own (possibly good) ideas on how to construct an asset allocation. While the 'robo' automation makes the relative complexity of the portfolio much more bearable.

jbolden1517
Posts: 868
Joined: Sun Jul 09, 2017 2:53 pm

Re: Backtesting some robo-advisers asset allocations

Post by jbolden1517 » Thu Aug 03, 2017 9:27 am

Well LadyGeek there you go. SIP with those adjustments now outperformed 3-fund by 204 basis points on average when backtested using your chosen methodology.

Forward tested I expect it to do better since that interval misses some critical advantages for inflation that this portfolio has. On the downside going forward value tiling was less popular then and so the advantages might diminish especially a few years after the SP500 index fund bubble bursts.

Post Reply